Security Impact Assessment Measures (SIAM)

SIAM
Security Impact Assessment
Measures
WP 3
Impact analysis on criminal actions
Deliverable D3.5
The
interdisciplinary
Research Group on
Law, Science,
Technology and
Society
State of the art report on SMT impact
on the global security level.
Report Title: CCTV and Smart CCTV
effectiveness: a meta-level analysis’
Vrije Universiteit
Brussel
1.Lucas Melgaço
2.Kristof Verfaillie
3.Mireille Hildebrandt
Project number
261826
Call (part) identifier
FP7-Security-2010-1
Funding scheme
Collaborative Project
Table of Contents
1. Executive Summary…………………………………..................................................... p. 03
2. Introduction………………………………………………………………………………………………. p. 04
3. Rationalisation, complexity and effectiveness……………………………………………. p. 06
4. Meta-analysis of the effectiveness of CCTV systems………………………………….. p. 11
5. From traditional CCTV to Smart CCTV………………………………………………………… p. 20
6. Conclusions……………………………………………………………………………………………….. p. 28
7. Bibliography………………………………………………………………………………………………. p. 31
2
1. Executive Summary
This deliverable contributes to the SIAM project by complementing the reports
based on empirical research in work package 3 (impact analysis on criminal actions).
The results of this report will contribute to the SIAM database and the Assessment
Support Tool.
The main objective of this deliverable is to present and discuss the complexities
involved in the assessment of the effectiveness of Security Measures and
Technologies (SMTs) in reducing crime and in increasing security. Although there is
an immense variety of SMTs currently operating, the analysis is limited to closedcircuit television (CCTV) and Smart CCTV because of the availability of relevant
literature on these two technologies. For most of the other SMTs such literature is
not accessible. This report focuses on texts which promote a meta-analysis of the
effectiveness of video surveillance, that is, articles that compile previous research on
effectiveness assessment or that promote methodological discussions about how to
better evaluate its effectiveness.
The first chapter, “Rationalisation, complexity and effectiveness”, is a theoretical and
abstract reflection about the assessment of effectiveness. It confronts the limits of
rationalisation processes with the complexities of crime. The chapter also discusses
the consequences of the current transition from a global focus on prevention to one
on pre-emption. Pre-emption, which can be seen as an attempt to rationalise the
unknown, an effort to predict the unpredictable, marks a new tendency in the way
security has lately been pursued.
The following chapter, “Meta-analysis of the effectiveness of CCTV systems”, dissects
a handful of reports on the evaluation of traditional CCTV systems. The chapter
enumerates the main difficulties present in CCTV evaluation and depicts the
mechanisms through which CCTV systems work. These mechanisms are classified in
terms of their “past”, “present” and “future” functions, a reasoning that can largely
be applied to other SMTs, particularly those that involve surveillance features.
The last chapter, “From traditional CCTV to Smart CCTV”, takes up the problem of
automation and video analytics. It brings in a host of issues around complexity,
discriminating between performance, evaluation, operation, policy concerns and
political considerations. This chapter promotes a discussion about the performance
of Facial Recognition Technologies (FRTs) and shows how the technology is – as yet –
not effective enough to prevent crime in complex settings.
The report ends by showing that statements on the effectiveness or ineffectiveness
of a certain SMT must be considered with caution, since such findings are contingent
upon a variety of constraints, such as temporal, spatial, and political. This report also
highlights that technical efficiency does not necessarily indicate crime-reducing
effectiveness, since the latter involves a higher level of complexity. Controlled
spaces, such as airports, train stations and especially parking lots, however, are
easier to manage than open systems. Therefore, in less complex settings surveillance
technologies seem to be more effective. Moreover, the less complex the context, the
easier and more precise one can assess the effectiveness of SMTs.
3
2. Introduction
Assessing effectiveness in science is by definition a complex task. This is already a
complicated effort in hard sciences like physics and biology and it can be ever more
obscure when it comes to social sciences. In the criminology field, for example,
discussions about whether a security measure is really effective in reducing crime
have always existed, not infrequently raising more contradictions than certitudes.
Such contradictions are in part due to the fact that sometimes reality can be just too
complex to be translated into a statistical model. However, this does not mean that
all attempts to measure effectiveness are worthless. Indeed, in some cases the
discussion about effectiveness seems to be unavoidable since, as Tilley (1999) points
out, evaluation and audit are imposed as major instruments of surveillance over
publicly funded programmes.
The main objective of this report is to present and discuss the complexities involved
in the assessment of the effectiveness of Security Measures and Technologies (SMTs)
in reducing crime, particularly those technologies used in transportation contexts
like airports, metro and train stations. Thus, the goal is not exactly to assert whether
a particular SMT is effective or not, but to analyse the complexities involved in such a
evaluation and to address some of the many variables that must be taken into
account.
Thus, this report should not be taken as a state of the art on the effectiveness of
SMTs. Firstly, because of the near impossibility of addressing all the available SMTs
used in mass transportation. The SMTs in question include technologies as different
as detection dogs to unmanned aerial vehicles. Moreover, criminal actions at
airports and in public transport systems encompass from everyday crimes to
terrorist attempts. Added to that it is the fact that new security technologies appear
every day, many of them being still very recent and almost no academic work about
their effectiveness has been published yet.
Among the different SMTs already in place in public mass transportation, closedcircuit television (CCTV) is one that has received more attention from scholars.
Despite the difficulties of precisely defining CCTV (for example, not all systems
classified under this label are necessarily “closed-circuit”; see IRISS (2012)), there is a
significant amount of work dedicated to discuss the effectiveness of such systems,
which justifies why this report is focused on this type of SMT. Less numerous but still
significant are the publications on automated CCTV systems, which include Video
Analytics and Facial Recognition features. Some of these publications will also be
treated here.
The second reason why this report is not entirely a state of the art is because the
literature discussed is not intended to be comprehensive. In the case of video
surveillance, for example, the amount of research and the number of publications
analysing effectiveness concerning specific case studies is considerable. This report
does not intend to consider all of them but rather to focus on those texts that
promote a meta-analysis of effectiveness, in other words, articles that bring together
4
previous research involving effectiveness assessment or that
methodological discussions about how to better evaluate effectiveness.
promote
The report is thus organized in three parts. The first part proposes a theoretical
reflection about rationalisation, complexity and the challenges involved in the
scientific discussions on effectiveness. The second part analyses the literature on the
effectiveness of traditional video surveillance systems and highlights the multiple
and complex variables that must be taken into account. The third part addresses a
discussion about automation by using the example of Smart CCTV and includes an
analysis about Facial Recognition applications. The conclusion shows how such a
framework can be expanded to the analysis of the effectiveness of other SMTs used
in public mass transportation, particularly those involving biometrics and profiling.
5
3. Rationalisation, complexity and effectiveness
The attempt to measure effectiveness can be understood as a rational practice. A
way to reduce our ignorance in relation to reality, and achieve a certain mastery of
it, is by the pursuit of reason, understood here as defined by Morin (2005, p. 94):
“Reason reflects the desire to have a coherent vision of phenomena, objects and the
universe. Reason has an undeniably logical aspect.” Rationality would thus be the
propensity to face the world from the aspect parting from reason. For Morin (2005,
p.94) “rationality is a game, the perpetual dialogue between our spirit, that creates
logical structures, implementing them to the world and which dialogues with this
real world.” Rationality refers to the desire to understand the world through a
scientific spirit. As a consequence, rationalisation is the process of applying
rationality to comprehend and the decoding of a given situation.
Ritzer (2011), through an interpretation of Weber’s concept of rationalisation,
suggests that a rational process involves four main principles: efficiency, calculability,
predictability and control. The author uses McDonald’s as an example to explain his
thesis. According to Ritzer, efficiency refers to the optimal method for accomplishing
a task. McDonald’s drive-through system would be a clear instance of the fastest
way to get from hungry to being full. The second principle, calculability, can be
noticed in the procedures currently present in several McDonald’s activities such as
the quantification of stocks and the calculus of the time it takes to deliver a product.
Predictability can be seen in the access customers have to similar products no matter
which McDonald’s they go to. Thus, predictability is related to routine and the
standardization of procedures. It is related to the quest for the reduction of risks and
unexpected situations. Finally, the existence of control in the McDonald’s
environment can be verified in actions like the standardization of employee’s
uniforms and in the close tracking of their tasks.
Considering rationalisation as a combination of efficiency, calculability, predictability
and control can be useful to the analysis of the effectiveness of SMTs. Assessing
effectiveness is, in itself, a rational procedure as it is an attempt to assert through
rational parameters whether a security measure is effective or not.
Not only actions, but also spaces can be rationalised (Santos, 1996). The
aforementioned example of McDonald’s drive-through exemplifies how space is
projected in a rational way to help costumers to go from a situation of hunger to that
of fullness without much effort. The spatial configuration of the drive-through
facilitates the quantification and control of both costumers and employees’ actions
in order to promote efficiency and predictability. In a similar reasoning, the quest for
security can in certain circumstances also be understood as a form of rationalisation
of space. Spaces may be designed or altered in order to reduce unpredictability and
fear. Rationalisation of space for purposes of security can occur in different manners.
For example, through the installation of surveillance cameras, body scanners,
construction of barriers, demarcation of limits, walls, through constructions that
regulate and restrict the movement of individuals and select those who have the
privilege to attend a particular place. Video surveillance, for example, can be seen as
an attempt to rationalise space as it includes the four principles suggested by Ritzer.
6
It permits calculability (the quantification of the number of persons or objects
present in a scene), efficiency (police officers may be better deployed), control
(intervention in actions that scape normality) and predictability (one may assume
that actions taken in a monitored space tend to be more predictable).
However, there are cases where rationalisation can be taken to an extreme. Ritzer
(2011) calls this “the irrationality of rationality”, a notion similar to that of
“rationalism” proposed by Morin (2005). While rationalisation is a process of
decoding the world - a necessary simplification of reality by which it becomes
tangible -, rationalism is the fact of aspiring to reduce reality to a purely rational
system. When the simplification is excessive, it denies the complexity of reality,
which could lead to biased interpretations. In this sense, rationalism would be the
desire to imprison reality into a coherent system, leaving aside everything that
eludes this logical scheme. Rationalism is thus an exaggeration of rationalisation
(Melgaço, 2012).
Rationalism is not uncommon in science. This is particularly true when policy makers
press scholars to simplify their advice in order to produce science-based decisions,
even in cases where knowledge in uncertain. As Stirling (2010, p. 1029) points out,
Expert advice is often thought most useful to policy when it is
presented as a single ‘definitive’ interpretation. Even when experts
acknowledge uncertainty, they tend to do so in ways that reduce
unknowns to measurable ‘risk’. In this way, policy-makers are
encouraged to pursue (and claim) ‘science-based’ decisions.
On this subject Morin (2005, p. 11) says:
While the simplifying thought disintegrates the complexity of the real,
the complex thought comprises as much as possible the simplifying
modes of thinking, but denies the mutilating, reductionist, onedimensional and finally blinding consequences of a simplification that
pretends to be the reflex of what is real in the reality.i
This reasoning proposed by Stirling and Morin can be transposed to the analysis of
effectiveness made in this report. The challenge remains exactly in addressing
assertive conclusions about the effectiveness of security measures without denying
the complexity of such question. One of the issues concerning the assessment of
effectiveness is related to the difficulties in considering all the possible variables. A
simplistic reasoning can conclude, for example, that since the attacks on the Twin
Towers on 9/11 in the United States no other event of the same magnitude has
taken place and that would be due to global investments in counter-terrorists
practices and technologies. However, the attacks might have ceased to take place
even without the expenses on security measures. The motivations and rationale of a
terrorist attack seem to be too complex to convey in a cause-effect scheme.
This justifies, in a certain sense, why Massumi (2007, p. 5) argues that we are moving
from a prevention to a pre-emption paradigm. He explains what prevention is by
saying:
7
Preemption is not prevention. Although the goal of both is to
neutralize threat, they fundamentally differ epistemologically and
ontologically. Epistemologically, prevention assumes an ability to
assess threats empirically and identify their causes. Once the causes
are identified, appropriate curative methods are sought to avoid their
realization. Prevention operates in an objectively knowable world in
which uncertainty is a function of a lack of information, and in which
events run a predictable, linear course from cause to effect.
Pre-emption, on the other hand, differs from prevention in that it acts on threats
that have not yet fully formed or that have not even emerged yet, “(i)n other words,
the threat is still indeterminately in potential” (Massumi, 2007, p. 13). Pre-emption
appears, thus, as a challenge of rationalizing the unknown, an effort of predicting the
unpredictable.
The recent investments in security initiatives, particularly those related to counterterrorist actions, are strongly influenced by the principle of pre-emption. The former
United States president George W. Bush, in a speech pronounced on June 2002
before the graduating class of the US Military Academy, affirmed (Massumi, 2007, p.
1):
If we wait for threats to fully materialize, we will have waited too
long. We must take the battle to the enemy, disrupt his plans and
confront the worst threats before they emerge. In the world we have
entered, the only path to safety is the path to action. And this nation
will act.
The concept of pre-emption then brings even more complexity to the discussion
about effectiveness. How can one assert that a certain technology is in fact effective
against a still unknown threat?
When it comes to measuring effectiveness, how to include all the variables involved?
How to avoid the lure of over-simplifying reality, which can lead to hasty
conclusions? Goodwin (2002) presents a case study that, if one analyses it
superficially, one could come to the possibly incorrect conclusion that the
installation of CCTV failed to deter crime. “Across the six camera locations covered in
the study, the number of crimes recorded by the police before the installation of
CCTV was 205, and the number of crimes recorded by the police after the installation
of CCTV was 213.” (Justice Analytical Services, 2009 p. 11). A more in-depth analysis
may, however, show that in fact the CCTV system increased the detection effect:
more crimes were detected but not necessarily more crimes were committed.
Farrington, Bennet and Welsh (2007) came up with similar findings in their work
comparing police recorded crime statistics with victimisation survey data in
Cambridge. They assert that in that case study CCTV surveillance led to the growth of
crime detection rates.
Another example of rationalisation involving security measures is that of cost
and benefit evaluation. Such evaluation involves calculability, control and
predictability of expenses and, overall, efficient trade-off between costs and
8
benefits. However, the complexity of the question makes cost and benefit evaluation
a very confusing issue. Previous reflections on the case of CCTV system can illustrate
such a complexityii. Armitage (2002, p. 3) highlights the multiple variables to be
taken into account when evaluating the costs of CCTV systems:
The cost of CCTV as a crime prevention measure includes not only the
initial investment but also the ongoing maintenance and running
costs. For this reason, any cost effectiveness analysis (as part of a
post-installation evaluation or a pre-installation feasibility study) must
account for these factors, in particular the staff time required to
monitor the cameras.
Tilley (1998, p. 149) adds that: “There will rarely if ever be sufficient data to assess
the full costs and benefits that can be directly attributable to CCTV. The potential
choice of what to include as a cost and a benefit is so wide that robust and
meaningful estimates will seldom be possible.”
Groombridge (2008, p. 74) argues that the benefits of CCTV do not compensate the
high costs. He declares: “My intention therefore is to argue from a radical
perspective – that is broadly sceptical of CCTV – that the Home Office, and therefore
the Treasury, has wasted enormous sums of tax payer’s money on the deployment
of CCTV.” Paradoxically, a report from the Scottish Government (Justice Analytical
Services, 2009, p. 26) seems to be more optimistic as it recognizes that CCTV footage
may bring financial benefits when used as evidence in court:
In light of the great expense of CCTV, its economic benefits,
particularly in relation to the savings gained through early
intervention and use of evidence in court, must be a focus of future
research. In order to gain a true picture of the associated economic
benefits, savings in these respects need to be balanced out against
the cost of installing and maintaining CCTV in the first place.
Thus, if assessing effectiveness is already a tough task, doing it through a precise
quantification of costs and benefits seems to be even more challenging.
It is equally difficult to translate abstract features like security, risk, and fear into
mathematical terms. Security, as Ceyhan (1998) states, is a complex and abstract
concept that involves both the absence of risks and the feeling of safety. Thus, how
to measure security? Should one measure it in terms of mathematical risks?
Moreover, how can one measure subjective features like the feeling of safety or
fear? One of the few essays on “measuring” fear is the work of Ditton (2000). The
author compared crime fear levels in Glasgow city before and after the installation of
CCTV and concluded that CCTV “is not making the unsafe feel safe, it is making the
already safe feel safer” (p. 702). Would it be possible to translate these findings into
precise mathematical terms? Would it be possible to measure security in terms of
the effectiveness of specific SMTs?
One should also keep in mind that statistics may “lie”, as argues Huff (1954) in his
book “How to lie with statistics”. The author depicts how intentional and
9
unintentional common errors associated with the interpretation of statistics can lead
to inaccurate conclusions. Statistics proving the effectiveness or the ineffectiveness
of a certain SMT may, in fact, be the result of manipulated or wrongly interpreted
data. In addition, Monmonier, in the book “How to lie with maps” (1996) adds that
maps, often used by criminologists to show the geographical development of
criminal patterns and to assert effectiveness, may similarly be object of
manipulation. Criminology studies have also addressed the difficulties of dealing
with statistics. Authors like Cicourel and Kitsuse (1963), Elliot (1995) and Maguire
(2002) point out the complexities and inaccuracies involved in the practice of
measuring and comparing crime rates.
In certain cases, evaluation reports have the ultimate goal of justifying costs.
Consequently, they can serve as a political tool. It is not incorrect to affirm that in
some occasions the final decision whether or not a certain SMT is effective ends up
being a political question rather than a technical one. As Tilley defends: “Sadly, many
really only want evaluations for self or political or organisational or civic
aggrandisement, even when purporting to want an independent piece or work. They
want the independent piece of work only if it comes up with the ‘right’ answer.”
(Tilley, 1998, p. 149). There are also cases where effectiveness is not even the main
purpose of the installation of SMTs. Policy makers sometimes install SMTs aiming to
promote the satisfaction of public demand for ‘something to be done’ about crime; a
demand that, as presented by Reiner (2002), has been inflated by the continual
representation of crime in the media discourse.
At this point it is important to distinguish efficiency from effectiveness. Some SMTs
may be efficient in relation to a specific goal they were designed to achieve, without
necessarily being effective in reducing crimes. While efficiency is related to
performing a task in an optimal way, effectiveness makes reference to choosing the
right tasks in order to achieve a goal. An iris scanner can be efficient in recognizing
individuals but may not be effective in reducing terrorist attacks in an airport, for
example.
Moreover, an approach that focuses too much on asserting the efficiency or the
effectiveness of a certain technology may ignore undesirable consequences caused
by the installation of such apparatus. For example, a scanner that allows an agent to
see intimate details of a passenger body may be a highly efficient technology for the
detection of explosives, but at the same time an intrusive initiative that raises clear
questions about privacy rights.
10
4. Meta-analysis of the effectiveness of CCTV systems
The first experiments involving the use of CCTV systems go back to the 1950s. These
systems were primarily applied to the management of public transportation and
later cameras were installed in banks and commercial centres. The pioneer country
to promote large investments in CCTV was the United Kingdom. In the 1970s, video
surveillance cameras were installed in four major British underground train stations
(Carli, 2008). It was only in the 1980s that CCTV was broadly expanded to the United
States and other European countries. During the 1990s, CCTV technology developed
significantly (IRISS, 2012) with the emergence of digital CCTV cameras and the
association of CCTV with the Internet.
Scientific publications about the effectiveness of CCTV systems started to appear in
the late 1970s. In 1979, Mayhew et al. (see also Burrows, 1979) published a study
about the effectiveness of CCTV cameras as part of a security effort to reduce crime
in some metro stations in London. The study found that cameras seem to be
effective in reducing the number of robberies, a finding also shared by Webb and
Laycock, in a later report published in 1992.
In 1999, Norris and Armstrong published “The Maximum Surveillance Society: The
Rise of CCTV”, one of the first books dedicated exclusively to the analysis of the
spread of video surveillance. Already in the introduction, the authors make it clear
they do not intend to present surveillance as merely a totalitarian initiative. There
are cases where surveillance conducted by the state (like knowing citizens’
addresses, income, and health conditions) is necessary to guarantee the protection
of the individual. On the other hand, as exposed by the authors, video surveillance
can reinforce prejudices by focusing differentially on the young, the male, the black
and the slovenly people.
One of the strengths of Norris and Armstrong’s book is the fact that the authors
carried out an empirical study of more than 600 hours of observation in three
different areas in the United Kingdom. The authors collected qualitative and
quantitative data of different agents involved and concluded that CCTV cameras are
not as effective in assisting the detection of criminal activities as normally suggested
by their proponents.
The compilation “Surveillance, Closed-Circuit Television and Control” edited by
Norris, Moran and Armstrong (1998) must also be highlighted. Four chapters of the
book are dedicated exclusively to the evaluation of CCTV. “Evaluating the
Effectiveness of CCTV Schemes”, by Nick Tilley, stands out because of its metaanalytical approach. According to the author, the text aspires “to help those with
responsibilities for CCTV evaluations better to think through the logic and rationale
of their work; and to highlight a number of technical problems that need to be
addressed in evaluating the effectiveness of CCTV schemes.” (p. 139).
Anyone trying to assess the effectiveness of CCTV systems, or any other SMT, must
have in mind the complexity of the task. Tilley (1998, p. 142-143) highlights such
complexity by enumerating a range of difficulties present in CCTV evaluation
procedures. These procedures were adapted and reinterpreted by us as follows:
11
i)
ii)
iii)
iv)
v)
vi)
vii)
viii)
ix)
Pseudo-random fluctuations in crime rates. In small areas crime can
‘naturally’ fluctuate independently of specific crime prevention efforts.
Regression to the mean. Crime prevention efforts are normally put in
place after periods of a high crime rate. However, crime rate may
naturally regress to a normal rate, therefore to the mean, even without
any particular intervention.
Floor effects. In contexts where crime rate is normally low, it is very
difficult to detect downwards effects as a result of CCTV.
Changes in background crime rates. The fluctuation crime rate of the
studied area must be compared to the development of crime in the
surrounding area.
Other changes in the area covered by CCTV. Researchers must take into
account other changes that might have happened in the studied area and
analyse how it may affect crime patterns or influence the effectiveness of
CCTV systems.
Changes in patterns of crime reporting and recording. CCTV can lead to
the increase of reporting crimes that would otherwise pass unnoticed.
CCTV as part of a package of crime prevention measures. The installation
of a CCTV system is normally accompanied by other measures that may
multiply the effectiveness of video surveillance technologies.
Displacement. In the case of CCTV actually being effective, crimes can
simply migrate to other less monitored areas or criminals can change the
time, method and type of crime. (see also Waples, Gill and Fisher, 2009).
Diffusion of benefits. As offenders are not fully aware of the CCTV
coverage they can change their behaviour even when acting in areas
outside the scope of the CCTV scheme.
When utilized for security purposes, surveillance cameras can be broadly classified
according to three principal mechanisms, which can be explained in terms of the
“past”, “present” and “future” functions of the criminal activity (Melgaço, 2012). In
relation to “past”, cameras have the intention to record events and serve as a data
bank for investigation and later identification of the criminal. The images can also be
used in court as evidence. In the relation to the “present” function, the camera has
the aim to serve as an extension of the eyes of the police or private security guards.
The agent behind the cameras identifies a suspected activity or a crime already in
the process of being committed and acts in real time, preventing it from being
accomplished. The third goal, which turns to “future” time, refers to the capacity of
the camera to prevent a crime from occurring by inducing a sensation into the
criminal that he is being continuously monitored.
Thus, a running camera, when connected to a system of data storage, responds to
“past”, “present” and “future” purposes as described. In other words, it permits
investigation, detection and deterrence. However, a connected camera, that does
not store images, only meets “present” and “future” functions. And finally, a false
camera in which images are neither produced nor stored, has only a function in the
“future”, since its sole purpose is to induce the feeling of being monitored.
12
Owen et al (2006), in their evaluation of the benefits of CCTV systems in managing
police resources, concluded that a considerable amount of police time could be
saved during the investigation process as a consequence of the use of CCTV. Hence,
it can be said that CCTV could be efficient in relation to the “past” function by
allowing police time to be used more productively.
The “present” function, that of immediately detecting crime or reacting to suspicious
behaviours, seems to be the one where cameras are less effective, particularly due
to the increasing amount of data to be monitored by agents. However, as will be
discussed later in this report, investments are being made in software dedicated to
the automatic detection of persons and behaviours.
Concerning the complexity involved in the “future” function, a report conducted by
the Police and Community Safety Directorate of the Scottish Government (Justice
Analytical Services, 2009, p. 23) states: “The issue of crime deterrence as a potential
outcome of CCTV is particularly difficult to evaluate as researchers are faced with the
problem of assessing crime that may have been observed, had CCTV not been
installed in the area.” Smith (2004 p. 377) is equally sceptical about the deterrence
effects of CCTV and believes that video surveillance is more effective as an
investigative tool:
I would argue that CCTV is generally ineffective as a crime prevention
tool. This is because the cameras, in these examples, are clearly not
producing “anticipatory conformity” in the population, deterring
criminals, nor are they offering Big Brother protection to those under
their gaze. Their use, in this type of scenario, is limited to the
reconstruction of events for post crime police enquiries.
When it comes to violent crimes, the literature (Sivarajasingham, Shepherd and
Matthews, 2003; Gill et al. 2006; Justice Analytical Services, 2009; Welsh and
Farrington, 2002) shows little evidence of a deterrent effect, which is
understandable due to the impulsive nature of this kind of crimes.
This framework of mechanisms by which a CCTV system works (“past”, “present”
and “future”) can be expanded as done by Armitage, Smyth, and Pease (1999, p.
226). According to the authors the means by which CCTV may prevent crime include:
 Caught in the act: perpetrators will be detected and possibly removed or
deterred.
 You’ve been framed: CCTV deters potential offenders who perceive an
elevated risk of apprehension.
 Nosy Parker: CCTV may lead more people to feel able to frequent the
surveilled places. This will increase the extent of natural surveillance by
newcomers, which may deter potential offenders.
 Effective deployment: CCTV directs security personnel to ambiguous
situations, which may head off their translation into crime.
 Publicity: CCTV could symbolize efforts to take crime seriously, and the
perception of those efforts may energize law-abiding citizens and/or
deter crime.
13
 Time for crime: CCTV may be perceived as reducing the time available to
commit crime, preventing those crimes that require extended time and
effort.
 Memory jogging: the presence of CCTV may induce people to take
elementary security precautions, such as locking their car, by jogging
their memory.
 Anticipated shaming: the presence of CCTV may induce people to take
elementary security precautions for fear that they will be shamed by
being shown on CCTV.
 Appeal to the cautious: cautious people migrate to the areas with CCTV to
shop, leave their cars, and so on. Their caution and security mindedness
reduce the risk.
Tilley (1998) also describes some of the mechanisms through which CCTV may work,
but he formulates it differently. According to him, CCTV systems:








Enable more effective deployment of security guards/police.
Increase ‘natural surveillance’ through increased usage of area by people
less fearful of crime.
Increase confidence of members of the public to intervene, if they believe
the situation is being observed and police back-up will follow.
Increase potential offenders’ fears that they will be seen, caught and
shamed of punished, or moved on for misbehavior or unwanted behavior.
Help catch offenders, who may then be removed.
Remind people to be cautious in areas covered, for instance where signs
tell them they are at risk.
Deter people from using the area because they deem it dangerous if CCTV
is needed.
Provoke offenders to search for an alternative area/situation in which
perceived risks of being seen are lessened.
Assessing the efficiency of cameras in fighting crime, Heilmann (2003) argues they
are more effective under specific conditions. According to the author, for a
monitoring program with cameras to be successful, one must consider: the use of
resources with technology potently sufficient to detect targets, the complexity of
urban space to be monitored, the correct definition of the targets and relevant goals
and the combination of other preventive measures.
Cusson (2005) identified that cameras do not have the same impact on all types of
crime, and that video surveillance has the best results: in the case of visible crimes,
in which offenders do not dare to confront their victims; in places that do not allow
criminals a quick escape; when the devices improve the capacity of identification and
intervention; when the installation of cameras is publicized and when monitoring
results in a constant intervention of security organs.
To date, a variety of case studies were conducted trying to assess the effectiveness
of CCTV systems. However, only few studies approached the methodological
difficulties of evaluating effectiveness, like in the case of Tilley (1998). Also rare are
14
comprehensive studies of the literature on the effectiveness of CCTV systems. In this
report we will highlight four studies that conducted a meta-analysis of other case
studies. The first is the research conducted by Welsh and Farrington (2002), who
compiled a systematic review of the literature up to that date. Searching by the
keywords “closed-circuit television”, “CCTV,” “cameras,” “social control,”
“surveillance,” and “formal surveillance”, the authors came up with forty-nine
evaluations to be analysed. Within this first survey they only included those
evaluations that met the following quality criteria:





CCTV was the main focus of the intervention;
There was an outcome measure of crime;
The evaluation design was of adequate or good methodological
quality, with the minimum design involving before-and-after
measures of crime in experimental and control areas;
There was at least one experimental area and one reasonably
comparable control area;
The total number of crimes in each area before the intervention
was at least twenty.
Out of the forty-nine previous evaluations, only twenty-two met the aforementioned
criteria for inclusion. The other twenty-seven evaluations were excluded because of
inconsistent methodologies. Ironically, they were not effective enough to be
considered for Welsh and Farrington’s effectiveness report. The main findings of this
compilation of papers were that:
CCTV had a significant desirable effect on crime, although the overall
reduction in crime was a rather small 4 percent. All nine studies
showing evidence of a desirable effect of CCTV on crime were carried
out in the United Kingdom. Conversely, the other nine studies
showing no evidence of any desirable effect of CCTV on crime
included all five North American studies. CCTV was most effective in
reducing crime in car parks. It had no effect on violent crimes but had
a significant desirable effect on vehicle crimes. (p. 110)
Four out of the twenty-two analysed cases deal with transportation. The four
researches were conducted in metro systems: one in the Montreal Metro by
Grandmaison and Tremblay (1997) and three in the London Underground, being two
by Webb and Laycock 1992) and one by Burrows (1979). According to Welsh and
Farrington the results show conflicting evidence of effectiveness:
[T]wo had a desirable effect, one had no effect, and one had an
undesirable effect on crime. However, for the two effective programs
in the London Underground, the use of other interventions makes it
difficult to say with certainty that it was CCTV that produced the
observed crime reductions, although in the program by Burrows
(1979), CCTV was more than likely the cause. Only two of the studies
measured diffusion of benefits or displacement, with one showing
evidence of diffusion and the other displacement. (p. 124)
15
Welsh and Farrington highlight the complexity of assessing effectiveness of CCTV
systems when other interventions like an increase in police patrol, improved lighting,
installation of passenger alarms, notices and signs about CCTV are included. They
argue that one-third of the twenty-two selected programs included interventions in
addition to CCTV, which “makes it difficult to isolate the independent effects of the
different components and the interactional effects of CCTV in combination with
other measures.”(p. 132). This reasoning can be expanded to other SMTs. It is
common for an SMT to be installed together with other technologies or practices,
which makes it sometimes difficult to isolate the effects of a single initiative.
Welsh and Farrington’s results, however, should not be immediately applied to the
present time since most of the studies they analysed were related to the
technological and political context of the 1990s. Since then, video surveillance
technologies have advanced drastically. Moreover, events like the bomb attacks on
the London underground in 2005 and particularly the attacks on the New York twin
towers in 2001 changed the way police, citizens and criminals deal with security
technologies and that include CCTV systems.
Three other comprehensives studies appeared after that. In 2005, Gill and Spriggs
published the results of a multi-evaluation of 14 English cities. One conclusion puts
forward that in only two out of fourteen cases the installation of CCTV systems
resulted in a reduced number of crimes. Moreover, in only one of the two successful
cases, which incidentally is a car park case, there was a significant reduction in crime.
Gill and Spriggs also found that recorded crime rates increased in several target
areas. This, as mentioned before, can lead to the wrong conclusion that CCTV failed
to effectively deter crime, when in fact it is possible that CCTV led to the increase of
crime detection. For this reason, Gill and Spriggs asserted that crime rates may be a
poor measure of the effectiveness of CCTV interventions.
In 2007, Welsh and Farrington prepared a report for the Swedish National Council
for Crime Prevention, which was later published as an article in 2009. The article is
an update of their first research conducted at the beginning of that decade. Welsh
and Farrington’s meta-analysis findings were consistent with those of their initial
review of 2002: “CCTV is most effective in reducing crime in car parks, is most
effective in reducing vehicle crimes, and is more effective in reducing crime in the UK
than in other countries.” (2009, p. 736). Their evaluations also showed that CCTV
schemes did not have a significant effect on crime in city and town centres, public
housing and public transport.
In 2009, the Justice Analytical Services of the Scottish Government prepared a report
which included only studies that had been conducted since the year 2000. The report
reviewed the evidence of the effect of CCTV on crime by looking at these four
parameters: crime deterrence effects; crime displacement effects; detection of
crime; and the use of evidence in the investigations and prosecutions process
(Justice Analytical Services, 2009). With the exception of crime displacement effects,
16
it can be said that the report comprises the aforementioned “past” (investigation),
“present” (detection) and “future” (deterrence) functions of video surveillance.
Due to a lack of case studies on the effectiveness of CCTV conducted since the year
2000, the report used selection criteria that were less strict than those used by
Welsh and Farrington (2002). The number of selected papers would have been too
small if they had used stricter criteria. In order to be selected, studies had to include
empirical efforts, CCTV had to be the main intervention included in the evaluation
and they had to have come up with an outcome measure of crime (Justice Analytical
Services, 2009, p. 6). Twelve studies were chosen, being five “quasi-experimental”
crime intervention studies, four interview studies, two studies incorporating
psychological experimental methods and theory and one observation study of
behavioural adaptations to CCTV.
Amongst others, the Scottish report analysed the work by Griffiths (2003), which
studied the effectiveness of a town centre CCTV system in Gillingham, UK. The
study’s findings include that video surveillance seems to be more effective within the
first year following the installation but such deterrence effects usually fade with
time. One of the reasons for this being the fact that media play a role in publicizing
the CCTV installation. Criminals aware of these security measures may feel
constrained in committing a crime. A very similar finding is presented by Armitage
(2002). She claims that the effectiveness of CCTV within London Underground
stations was reduced after approximately one year and that other CCTV evaluations
revealed that the initial reductions in crime following the installation of CCTV can
diminish if publicity is not maintained. In other words, it can be said that while the
“past” and the “present” functions of video surveillance may keep up the same
degree of effectiveness over time, the “future” function has an expiry date.
The evaluation conducted by Sivarajasingham, Shepherd and Matthews (2003) was
also highlighted by the Scottish report. This study addressed the aforementioned
common confusion between real changes in crime rates, and increases in crime
detection and recording. Furthermore, the authors focused on the effects of CCTV on
violent crime and, different from previous research, suggested that CCTV may have a
desirable effect on these type of crimes. Increased detection would lead to more
punishment, which would deter criminals from committing violent acts (Justice
Analytical Services, 2009).
Concerning the evaluation of CCTV effectiveness in violent crimes, the strategy
adopted by Gill et al (2006) is noteworthy. The authors led focus group discussions
with ten murderers, who were about five years into their sentence, and seven had
committed the murder in a public place. “The aim of this study was therefore, to
investigate whether the murderers involved in the focus groups would have been
deterred by the presence of CCTV cameras at the time of committing their offence.”
(Justice Analytical Services, 2009, p. 11). The results of this research are different
from those found by Sivarajasingham, Shepherd and Matthews (2003). Gill et al
found evidence that the murderers did not consider the presence of cameras as an
important deterrent technology. Moreover, when the crimes were committed under
the influence of alcohol, they ignored CCTV cameras even more.
17
According to Tilley, the existence of conflicting results in evaluations of the
effectiveness of CCTV systems, like these presented by Sivarajasingham, Shepherd
and Matthews (2003) and Gill et al (2006), was to be expected:
So, though mixed findings from evaluation studies can often be
expected simply because many pieces of work are technically flawed
or technically different, they are also inevitable because measures will
have differing impacts depending on the conditions in which they are
introduced. This means that the frequently asked question, ‘Does
CCTV work?’ admits and can admit of no consistent answer. It is,
therefore, not a sensible, useful or intelligible question to address,
notwithstanding the frequency with which it is asked or the money
and effort spent trying to answer it. (Tilley, 1998, p. 144).
Thus, as said in the Scottish report: “The ‘effectiveness’ of CCTV must be considered
in light of its intended purpose, as each individual project is installed to serve its own
purpose.” (Justice Analytical Services, 2009, p. 23).
Despite these conflictive findings, there seems to be at least one consensus: CCTV
systems appear to be significantly effective in reducing crimes in less complex
settings, such as parking lots (Webb and Laycock, 1992; Armitage, 2002; Griffiths,
2003; Carli, 2008; Welsh and Farrington, 2009). This finding is consistent with the
reasoning outlined in the introduction of this report: the more complex a situation or
a space is, the larger the number of variables to be considered; hence, more
challenging the rationalisation process. In 1992, Webb and Laycock had already
alerted that “CCTV does not seem very useful in large, complex and crowded
environments to deal with more surreptitious behaviour such as pickpocketing or
shoplifting.” (1992, p. 23). Similar findings were produced by Ditton and Short (1999)
in their study about the effectiveness of open street CCTV in two adjacent British
town centres: Airdrie, a small town, and Glasgow, a large city. The authors indicated
that CCTV seemed to be more effective in the first case. The Scottish report puts
forward a similar reasoning when it says that “Recent research has resulted in
evidence consistent with the repeated finding that CCTV may be more effective in
deterring crime in smaller and less complex areas than large city centres.” (Justice
Analytical Services, 2009, p. 2). It can thus be said that the less complex a situation or
a space is, the more effective an SMT may be in reducing crime, and the more
precise the evaluation of such effectiveness will be.
Complex settings, on the other hand, present themselves as a challenge to
rationalisation. The multiplicity of variables to be considered makes it complicated to
translate them into terms of efficiency, calculability, predictability and control.
Moreover, considering the three functions of CCTV – “past”, “present” and “future”
– the “present” function seems to be the least effective in complex settings.
Detecting suspicious activity, and acting upon it before it becomes an accomplished
crime, is an enormous challenge to guards and police officers acting in intricate
contexts.
One way to increase effectiveness in relation to the “present” effect is by setting up
CCTV control rooms, where agents can manage multiple cameras from and monitor
18
several screens at a time. Smith (2004) conducted field research in the CCTV control
room of a British college. Within the findings of his ethnological research is what he
named the “boredom factor”. In such a context, like a college, the occurrence of
criminal activities is significantly low. The agents, thus, get bored and watch the
screens with less attention and interest. Moreover, he found that “the operatives
felt alienated from their job, due to the imprisoning confines of the CCTV control
room, the long hours worked, the high expectation levels placed upon them and the
low pay and lack of acclamation received from their employers” (p. 376). The result
is the inefficiency of the system in flagging crime in real time.
Moreover, with the spread of CCTV cameras the number of images produced
steadily grows. This means that more and more images are to be analysed. Besides
the installation of CCTV control rooms, another response to this increasing flow of
information has been the use of software that is able to automatically detect faces
and suspicious activity. This topic will be discussed in the following part of this
report.
19
5. From traditional CCTV to Smart CCTV
Both the increasing capacity of data recording and the widespread use of cameras
have resulted in an immense volume of data at hand. It is important to highlight that
although the spread of CCTV cameras has lead to an overload of data being
gathered, it did not necessarily lead to an increase in informationiii. A traditional
CCTV camera merely produces a set of pixels, an amount of data to be interpreted
and transformed into information. Footage is traditionally interpreted by human
operators, who, by watching computer screens, decide whether or not an event is
normal or abnormal. Humans, however, are exposed to failures due to reasons like
those mentioned by Smith (2004). In response, software has been developed to
enable facial recognition and automatic detection of suspicious activities and
behaviour. As Adams and Ferryman (2012, p. 2) argue: “The rapid proliferation in the
number of CCTV installations worldwide, in areas such as shopping centres,
underground stations, and airports, has expedited the demand for automatic
methods of processing their output.” The demand for automated surveillance is also
presented in the military context, where drones create large data bulks to be
processed and analysed.
Usually called Smart CCTV, Intelligent CCTV, or Video Analytics, these automated
systems differ significantly from traditional analogue video surveillance in respect to
their possibilities. What makes them smart is the combination of images and
software. Ferenbok and Clement (2012) highlight the following in their definition of
video analytics:
Video analytics (VA) is software that uses signal processing and
pattern recognition techniques to automatically generate meaningful
or semantic data from video images. Video analytics marks a
paradigmatic shift in visual surveillance practices — in how
information is purposed, and repurposed — and in the potential
consequences for surveillance subjects.
Adams and Ferryman (2012, p. 3) put forward that “The overall capability to
automatically analyse video images to extract objects, detect events, and to perform
behavioural analysis, is referred to as video analytics.”
Detection, tracking and classification of targets are the main tasks expected from
video analytics engines. Detection refers to the identification of what physical
objects exist in the surveillance area, tracking is the understanding of how they move
and classification is related to the labelling of objects as human, vehicle, animal and
to the interpretation as normal or abnormal objects or behaviours.
In video analytics, data can be interpreted from a single frame from one camera,
across a video sequence from one camera, or in comparing different views of the
same area from multiple cameras (Adams and Ferryman, 2012, p. 3). Data can also
be interpreted from cameras in different areas, as in the case of transport
applications where inferences on speed can be made through the identification of
the same object in two different spots (Rios-Cabrera, Tuytelaars, and Van Gool,
2012).
20
As pointed out by Haering, Venetianer, and Lipton (2008, p. 281), the first
applications of automated video surveillance systems were related to simple motion
detection. Cameras could, for example, be set to start recording only and as soon as
they detected movement. The usefulness of motion detection systems was,
however, reduced due to the high false alarm rates caused by the movements of
unimportant objects like shadows, foliage, or small animals.
The following step in the evolution of intelligent CCTV was the creation of object
based video analysis. This type of analysis reduced the occurrence of false alarms by
introducing sophisticated filtering capabilities. Alarms would sound only in the case
of movement of specific types of objects.
The detection of object functions as follows: “Pixels deviating from the background
model statistics are labeled as foreground. These pixels are grouped together into
spatial blobs, then tracked, thus creating spatiotemporal objects” (Haering;
Venetianer; Lipton, 2008, p. 281). The distinction between objects and background is
also influenced by the quality of lighting of the area. The more complex the
background and the worse the lightning, the less precise the identification of objects
will be. As Adams and Ferryman (2012, p. 3) pointed out, object identification in
small and indoor scenes tends to be simpler as ambient lighting is more controlled
while applicability to dense, crowded environments is much more limited. Dee and
Velastin (2008) assert that Smart CCTV engines perform with decreased performance
in unstructured or changing environments such as public places.
Although one of the main expected applications of video analytics is the
improvement of the “present” function of surveillance, Adams and Ferryman (2012)
argue that most of its applications still do not operate in real time, being currently
used for after-the-fact investigation. Ferenbok and Clement (2012) also address this
discussion when they say the following:
Video Analytics (VA) addresses at least two major limitations of the
conventional analog CCTV model: live monitoring and retrospective
searching. Watching video surveillance can be tedious and boring.
Often there can be hours or days of video from multiple sources
where very little of interest actually happens. The volume of
information produced by multiple cameras running 24 hours seven
days a week means that much of the information captured by analog
CCTV cameras is not viewed in real-time or retained, and if recorded,
remains effectively not viewed.
Significant effort, however, has been put into increasing the speed of object
detection. Lately, a lot of attention has been going to automatic number plate
recognition (ANPR) – also called License Plate Reader/License Plate Recognition (LPR)
– as well as to the automatic detection of abandoned bags. As presented by Carli
(2008), ANPR is a surveillance technology that uses optical character recognition of
images to read license plates. When acting in real-time, ANPR can help the police
chasing a vehicle, as well as for monitoring traffic activity. ANPR can thus be
considered an example of rationalisation of spaces: streets equipped with CCTV
cameras connected to ANPR systems are spaces that permit the quantification and
21
the control of traffic activity and an increase in the efficiency of police activities.
The development of faster real-time responses is particularly desirable in the
detection of abandoned bags. The longer it takes to detect such objects, the higher
the risks. The European Union recently concluded a project called SUBITO
(www.subito-project.eu) “aimed at developing a surveillance system for robustly
detecting abandoned bags in public spaces and to identify and track the owner”
(Adams and Ferryman, 2012, p. 6). As highlighted by the project, detecting
abandoned bags automatically is a complex task. An example of a criterion normally
taken into account by the algorithms is when the supposed owner of a bag is, for a
certain time, further than a given distance from the bag. However, it is possible that
in a busy airport people move closely enough to a bag, which may confuse the
system in classifying it as unattended. As Adams and Ferryman (2012, p. 6) point out:
“For dealing with these more complex scenarios, it became necessary to derive a
more complete activity analysis and the concept of social groups was introduced.”
The SUBITO project developed a framework to automatically understand the
behaviour of groups, which helped to reduce the number of false alarms for
abandoned bags.
The example of applying video analytics in detecting abandoned bags gives proof of
the increasing capacity of software to translate social relations into computational
language. Concepts like ownership of an object and belonging to a social group are
only two of a variety of examples of social relations decoded in rational and
mathematical models.
A similar process of digitisation and quantification is currently happening with the
human body. Body features like the iris, face and hands are being translated into
mathematical formulas in a process commonly named biometrics. In the case of
Smart CCTV, the most evident biometrical application is in the use of cameras for
facial recognition.
As explained by Carli (2008, p. 5), “Facial recognition technology is a computer
application for automatically identifying or verifying a person from a digital image or
video frame from a video source.” One of the advantages of facial recognition
technologies (FRTs) in comparison to other biometric systems is that, as pointed out
by Introna and Wood (2004), they can operate anonymously in the background,
while iris or hand scanners require a direct physical involvement from their targets.
Moreover, FRTs are relatively inexpensive and can, in principle, take advantage of
existent traditional CCTV systems, as FRTs are more software than hardware-based.
Video analytics is a still recent technology – the first experiments with facial
recognition technologies date from the late 1990s – which explains why very few
case studies have been published about their effectiveness in reducing crime. Even
more rare are the meta-level studies about Smart CCTV effectiveness along the lines
of the four reports on traditional CCTV mentioned before. Three of the few
exceptions are the aforementioned work of Adams and Ferryman (2012), Dee and
Velastin (2008) and the report produced by Introna and Nissenbaum (2009) on what
could be called an evaluation of other evaluations on FRTs.
22
According to Introna and Nissenbaum, there already exist a considerable amount of
publications on the technical side of FRTs and some studies on the social impact of
these systems but no publication connecting both:
On the one side, there is a huge technical literature on algorithm
development, grand challenges, vendor tests, etc., that talks in detail
about the technical capabilities and features of FRT but does not
really connect well with the challenges of real world installations,
actual user requirements, or the background considerations that are
relevant to situations in which these systems are embedded (social
expectations, conventions, goals, etc.). On the other side, there is
what one might describe as the “soft” social science literature of
policy makers, media scholars, ethicists, privacy advocates, etc., which
talks quite generally about biometrics and FRT, outlining the potential
socio-political dangers of the technology. This literature often fails to
get into relevant technical details and often takes for granted that the
goals of biometrics and FRT are both achievable and largely Orwellian.
Bridging these two literatures—indeed, points of view—is very
important as FRT increasingly moves from the research laboratory
into the world of socio- political concerns and practices (Introna and
Nissenbaum, 2009, p. 8)
Through an approach that is in line with Morin’s aforementioned reflections on the
idea of complexity, their report “attempts to straddle the technical and the sociopolitical points of view without oversimplifying either” (p. 8).
The authors analyse FRTs through five main parameters (pp. 3-5), of which the first
three are of special interest to our discussion:
1. Performance: What types of tasks can current FRT successfully
perform, and under what conditions?;
2. Evaluations: How are evaluations reported? How should results be
interpreted? How might evaluation procedures be revised to produce
more useful and transparent results?;
3. Operation: What decisions must be made when deciding to adopt,
install, operate, and maintain FRT?;
4. Policy concerns: What policies should guide the implementation,
operation, and maintenance of FRT?;
5. Moral and political considerations: What are the major moral and
political issues that should be considered in the decision to adopt,
implement, and operate FRT?
The literature analysed by Introna and Nissenbaum shows that in general FRT
performance has proven effective in the case of relatively small populations in
controlled environments. Consequently, facial recognition is much more efficient in
applications of verification than identification. As the authors explain, verification
23
consists of matching an individual’s face to a pre-existing image “on-file” associated
with the claimed identity. Identification, on the other hand, refers to recognizing
individuals who did not voluntarily offer their biometrics. The literature shows that
in these more complex cases where FRTs are used to match an individual’s face with
any possible image “on-file”, the performance results are much poorer.
There are two types of identification: closed-set and open-set. In closed-set
identification, it is known in advance that the wanted individual’s profile is already
present in the database. Whilst in more complex open-set situations, it is not known
in advance whether an individual is present or not in the gallery image. As Introna
and Nissenbaum (2009, p. 12) stress:
The outcome of these two identification problems will be interpreted
differently. If there is no match in the closed-set identification then
we know the system has made a mistake (i.e., identification has failed
(a false negative)). However in the open-set problem we do not know
whether the system made a mistake or whether the identity is simply
not in the reference database in the first instance.
Most of real-world identification applications are open-set, which explains the poor
performance of FRTs in more complex situations. As a consequence, the authors
affirm: “the ‘face in the crowd’ scenario, in which a face is picked out from a crowd
in an uncontrolled environment, is unlikely to become an operational reality for the
foreseeable future.” (Introna and Nissenbaum, 2009, p. 3)
The performance of verification and, especially, identification tasks depends on a
series of factors. One of the most important aspects is the quality of the gallery
image used to identify a certain individual. In the case of a verification task, the
conditions of enrolment are ideal, as individuals opt to voluntarily offer their
biometric information. In this case, the image captured by the camera is comparable
to passport quality photographs. However, in identification cases this often is not the
case since the gallery is composed of low quality images. Charlie Savage, in an article
for the New York Times (Facial Scanning is Making Gains in Surveillance – August 21,
2013), proposes a similar reasoning:
The automated matching of close-up photographs has improved
greatly in recent years, and companies like Facebook have
experimented with it using still pictures. But even with advances in
computer power, the technical hurdles involving crowd scans from a
distance have proved to be far more challenging. Despite occasional
much-hyped tests, including one as far back as the 2001 Super Bowl,
technical specialists say crowd scanning is still too slow and
unreliable.
The literature analysed by Introna and Nissenbaum (2009, p. 3) showed that
performance is also contingent of other factors like:
Environment: The more similar the environments of the images to be
compared (background, lighting conditions, camera distance, and thus
24
the size and orientation of the head), the better the FRT will perform.
Image Age: The less time that has elapsed between the images to be
compared, the better the FRT will perform.
Consistent Camera Use: The more similar the optical characteristics of the
camera used for the enrollment process and for obtaining the on-site
image (light intensity, focal length, color balance, etc.), the better the
FRT will perform.
Gallery Size: Given that the number of possible images that enter the
gallery as near-identical mathematical representations (biometric
doubles) increases as the size of the gallery increases, restricting the
size of the gallery in “open set” identification applications (such as
watch list applications) may help maintain the integrity of the system
and increase overall performance.
In regard to performance, Introna and Nissenbaum (2009) also analysed the
conditions that may limit the efficiency of FRT, in other words, “what makes it not
work” (p. 38). The authors highlight that FRT effectiveness is a consequence of the
performance of the whole operational system rather than of a particular
technological issue. In addition, they point out that “The successful operation of a
FRS [facial recognition system] in the identification mode is critically dependent on
the key characteristics of the gallery database: image quality, size, and age.” (p. 39).
It is also important to take the quality of the probe image and the conditions of
capture into account. Best FRT performance occurs when the conditions under which
the probe photos were taken most closely resemble those of the gallery image.
Moreover, the recognition algorithms should be carefully chosen since different
algorithms can produce different results.
Concerning the evaluations of FRT, Introna and Nissenbaum divide them into three
categories: technological, scenario and operational. Technological evaluations
address the capabilities of algorithms to promote facial recognition under closely
controlled conditions. Scenario evaluations deal with the capacity of the system to
recognise faces in a specific scenario designed to emulate a real-world situation.
Finally, operational evaluations are related to the application of FRT to real in-locus
situations.
The authors analysed publications in these three different categories and came to
the conclusion that the best results were found in technological evaluations. Thus,
the more complex the settings are, like in real operational contexts, the less efficient
an FRT is. It is not uncommon though for positive findings of technological
evaluations to be used as a justification for investment in FRT in real operational
situations. Introna and Nissenbaum (2009, p. 4) affirm that:
Evaluation results must be read with careful attention to pre-existing
correlations between the images used to develop and train the FRT
algorithm and the images that are then used to evaluate the FRT
algorithm and system. Tightly correlated training (or gallery) and
25
evaluation data could artificially inflate the results of performance
evaluations.
In regard to the operation of FRT, Introna and Nissenbaum highlight how delicate the
balance is between the level of certainty to which a system works and the
acceptance rates. There is a trade-off between False Acceptance Rates (FAR; the
probability that a system incorrectly matches the captured biometric feature with
the stored template, creating a false positive) and False Rejection Rates (FRR; the
probability that the system fails to detect a match, creating a false negative). They
explain (2009, p. 4):
For instance, a system with a high threshold, which demands a high
similarity score to establish credible recognition in the verification
task, would decrease the number of individuals who slip past the
system (false accept mistakes), but would also increase the number of
individuals who would be incorrectly rejected (false reject mistakes).
These trade-offs must be determined, with a clear sense of how to
deal with the inevitable false rejections and acceptances.
When applying FRT to verification issues, it is common to set a high threshold for the
system. In such a situation, it is crucial to guarantee a low quantity of false accept
mistakes, even at the expense of a high number of false reject mistakes. However,
when FRT is applied to identification procedures, a high rate of false alarms may
tamper the applicability of such technologies. False alarms require extra resources
for constant follow-up. Moreover, the repetition of constant false alarms may lead
operators to ignore a real alarm.
Introna and Wood (2004), Introna and Nissenbaum (2009) and Adams and Ferryman
(2012) draw attention to two scenario applications where facial recognition systems
were abandoned due to their lack of utility. In Tampa Bay, in the US, this happened
because the system generated a large number of false positive alarms. In the Palm
Beach Airport case, a group of 15 volunteers were compared to a small database of
only 250 images and the results of a mere 47% of correct identifications motivated
the abandonment of the project.
False positive alarms are not just technical issues; they are also political. Operators
may, for example, deal differently with a wrongly target individual, who is a member
of a minority group. Moreover, as Introna and Wood (2004, p. 192) asserts “The
operators may even override their own judgments as they may think that the system
‘sees something’ that they do not.” Situations like these could be labelled as cases of
rationalism, where, as termed by Ritzer (2011), the process of rationalisation
resulted in the “irrationality of rationality.”
Introna and Nissenbaum (2009) conclude their report by saying that the analysed
evaluations suggest that FRT can be useful in verification task as long as certain
conditions are met. However, so far their performance has still been very poor in
process of identification, whether closed- or open-set. Adams and Ferryman (2012,
p. 9) reinforce this idea by saying that the current automated surveillance systems
operate with satisfactory performance in restricted domains in which algorithms
26
function well. “However, the overall vision is to develop systems which can respond
to and act in the real world.” Thus, facial recognition systems, and more broadly
video analytics, are technologies that are efficient in controlled settings but still very
inefficient in real complex situations.
27
5. Conclusions
Anyone involved in assessing the effectiveness of Security Measure Technologies
must keep in mind that this is not a simple task. As this report showed, the
discussion about effectiveness is full of complexities of all sorts. Asserting the
effectiveness of a technology in reducing crime is both a technical and a political
task.
Although there is an immense variety of SMTs currently operating in mass
transportation, the choice for CCTV and Smart CCTV was based on the availability of
the bibliography on these two technologies. The number of case studies on the
effectiveness of video surveillance systems abounds. This plentiful offer of
references allowed authors like Welsh and Farrington (2002; 2007), Gill and Spriggs
(2005), and Justice Analytical Services (2009) to produce comprehensive reports on
the state of the art of such technology. In these four reports, the reference to the
notion of complexity is evident. The work of Tilley (1998) must also be highlighted.
Despite not being a compilation like the other reports, it addresses crucial
methodological questions about the difficulties involved in assessing the
effectiveness of video surveillance systems.
The inclusion of Smart CCTV in our analysis is justified by two reasons. Firstly,
because Smart CCTV seems to be an unavoidable update to traditional CCTV
systems. These systems are becoming more “intelligent” and automated each day.
Not taking this development into account may thus result in an out-dated
evaluation. Secondly, the logic behind the functioning of video analytics systems
shares similarities to that currently put in place with other SMTs. The understanding
of the reasons why facial recognition systems may fail can be used to help analysing
other technologies like body scanners, for example.
There exist, however, fewer publications on Smart CCTV than on traditional CCTV.
The number or existent case studies is considerably lower, and methodological and
comprehensive state of the art reports are very rare. Nevertheless, the works of
Adams and Ferryman (2012) and Dee and Velastin (2008) and particularly the report
prepared by Introna and Nissenbaum (2009), were highlighted because of their
comprehensive approach to the discussion and because of how they underscore the
complexity of the question.
Reducing crime, measuring the reduction of crime, asserting that this reduction was
due to the effects of a certain technology, and measuring such effects are four
extremely complicated issues. The reality is so complex that a policy maker may be
surprised by the fact that an SMT can actually augment crime instead of diminishing
it. As Welsh and Farrington (2003, p. 111) explain: “The presence of CCTV may give
people a false sense of security and cause them to stop taking precautions that they
would have taken in the absence of this intervention, such as not wearing jewelry or
walking in groups when out at night.” Consequently, the rationalisation of the
question through a cause-effect analysis will be insufficient to apprehend all the
intricacies involved.
28
As presented, video surveillance, whether traditional or “smart”, functions through a
series of mechanisms that can be classified into three main effects: “past”, “present”
and “future”. The “past” effect, that is to say, the capacity of video surveillance to
serve investigation purposes appeared as the one in which both CCTV and Smart
CCTV are more effective in reducing crime. The growing digitisation of video
surveillance and the increasing capacity of saving data has lead to what MayerSchönberger (2009) named the almost impossibility of forgetting in the digital age.
This phenomenon improved the “past” effect of video surveillance since the amount
of information recorded and stored in CCTV databases grows each day. In regards to
the “present” function, that of detecting crimes in real time and acting upon it, CCTV
seems to be very ineffective, whilst Smart CCTV showed better results. Concerning
the “future” aspect, that is, deterrence, video surveillance appeared more effective
in the first year after the installation of cameras. However, such effectiveness tends
to decrease over time.
Another important finding extracted from the report analysis is that the concept of
effectiveness is often used as a political discourse. In fact, there are cases where
technical assessments of the efficiency of certain SMTs are confounded with their
real effectiveness in reducing crime. This is currently the case with facial recognition
systems, the technological evaluations of which – that is, the capabilities of
algorithms to promote facial recognition under closely controlled conditions – are
mistaken for its real operational capabilities. In other words, the positive findings
about the efficiency of facial recognition engines are used as a justification for
implementing this technology in real operational situations of combating crime.
Efficiency is, thus, mistaken for effectiveness.
Some findings regarding the functioning of facial recognition technologies (FRTs) can
be expanded to biometrics as a whole. Whether it is a fingerprint, iris or body
scanner, the logic involved in the trade-off between false acceptance rate (FAR) and
false rejection rate (FRR) is always present. The decision on the FAR and FRR
depends on the application under consideration and has important social and
political implications.
The analysis of the effectiveness of FRT also pointed out the importance of
databases. The main goal of an FRT is to match faces with profiles on an image
gallery. As shown, effectiveness is a function of the quality of this gallery. The quality
of such database depends on the time lapse between the gallery image and the
captured probe image. Consequently, it can be said that FRT efficiency decreases
with time, which draws attention to the importance of keeping a database always
updated.
Finally, the most important finding of this meta-level analysis goes back to the
discussion about complexity mentioned at the outset. The majority of the analysed
bibliography mentioned that video surveillance, whether traditional or smart, is
more effective in simpler and controlled settings, while performance declines as the
context gets more complex. Traditional CCTV seems to be effective in the
investigation, detection, and deterrence of crime in parking lots. These are very
simple contexts where the number of involved variables is reduced.
29
Smart CCTV uses the same logic. The literature showed that FRT tends to be much
more efficient in verification than identification procedures, which is explained by
the fact that the former is more controlled while the latter involves a higher number
of variables. Introna and Nissenbaum (2009, p. 24) also indicate the connection
between complexity, the number of involved variables and effectiveness of FRT by
saying:
To conclude this discussion, we can imagine a very plausible scenario
where we have a large database, less than ideal images due to factors
such as variable illumination, outdoor conditions, poor camera angle,
etc., and relatively old gallery images. Under these conditions,
performance would be very low, unless one were to set the FMR
[false match rate] to a much higher lever, which would increase the
risk that a high number of individuals would be unnecessarily
subjected to scrutiny.
The relationship between complexity and effectiveness can be expanded to a
general analysis of SMTs in public mass transportation. An airport, for example, can
be considered a more controlled setting than a train station. In an airport, the public
is selected and space is rationalised through the implementation of checkpoints used
to quantify and control the flux of passengers. On the other hand, a train station
looks more like an open system where the input of variables is more complex
compared to an airport. According to this reasoning, it is expected that SMTs
implemented in airports would be more effective than those installed in more
complex and unpredictable contexts.
In conclusion, the assessment of the effectiveness of CCTV and Smart CCTV in
reducing crime must inevitably take into account all the complexities involved in
such question. Therefore, this is not a mere cause-effect question to be answered by
technicians but a multifaceted issue to be addressed by a combination of different
specialties.
30
7. Bibliography
Ackoff, R. L. (1989) "From Data to Wisdom", Journal of Applied Systems Analysis,
Volume 16, 3-9.
Adams, A. A. and Ferryman, J. M. (2012). The Future of Video Analytics for
Surveillance and its Ethical Implications (November 12). Security Journal,
Forthcoming. Available at: http://ssrn.com/abstract=2174255. Accessed October 8,
2013.
Armitage, R. (2002) To CCTV or not to CCTV?: A review of current research into the
effectiveness of CCTV systems in reducing crime. Nacro Briefing Note. Available at
http://epic.org/privacy/surveillance/spotlight/0505/nacro02.pdf. Accessed October
8, 2013.
Armitage, R., Smyth, G., and Pease, K. (1999). Burnley CCTV evaluation. In K. Painter
and N. Tilley (eds.), Surveillance of Public Space: CCTV, Street Lighting and Crime
Prevention: Vol. 10. Crime Prevention Studies, pp. 225-50. Monsey, NY: Criminal
Justice Press.
Burrows, J. N. (1979). The impact of closed circuit television on crime in the London
Underground. In Crime in public view, Home Office Research Study no. 49, edited by
Patricia Mayhew, Ronald V. G. Clarke, John N. Burrows, J. Michael Hough, and Simon
W. C. Winchester. London: HMSO.
Carli, V. (2008) Assessing CCTV as an effective safety and management tool for
crime-solving, prevention and reduction. Comparative Analyses Report. Montreal:
International Centre for the Prevention of Crime. Available at: http://www.crimeprevention-intl.org/fileadmin/user_upload/Publications/La_videosurveillance_estelle_un_outil_de_securite_et_de_gestion_efficace_FR.pdf. Accessed October 8,
2013.
Ceyhan, A. (1998) Analyser la sécurité: Dillon, Waever, Williams et les autres.
Cultures and Conflits, 31-32. Available at: http://conflits.revues.org/index541.html.
Accessed October 8, 2013.
Cicourel, A. V. and Kitsuse, J. I. (1963): A Note on the Use of Social Statistics. Social
Problems, 11, 131-139.
Cusson, M. (2005). La surveillance et la télésurveillance: sont-elles efficaces? Revue
internationale de criminologie et de police technique et scientifique, vol. 58, no 2,
avril-juin
2005,
131-150.
Available
at:
http://classiques.uqac.ca/contemporains/cusson_maurice/surveillances_et_telesurv
eillances/surveillances_et_telesurveillances.pdf. Accessed October 8, 2013.
Dee, H. M. and Velastin, S. A. (2008). How Close Are We To Solving the Problem of
Automated Visual Surveillance? A Review of Real-World Surveillance, Scientific
Progress and Evaluative Mechanisms. Machine Vision and Applications, 19(5-6), 329343. Available at: http://www.comp.leeds.ac.uk/hannah/papers/mva07.pdf.
Accessed October 8, 2013.
31
Ditton, J. (2000). Crime and the city: Public attitudes towards open-street CCTV in
Glasgow. British Journal of Criminology, 40, pp. 692 – 709.
Ditton, J. and E. Short (1999) Yes, it works, no it doesn't: Comparing the effects of
open-street CCTV in two adjacent Scottish Town Centres, Crime Prevention Studies,
10,
pp.
201-223.
Available
at:
http://www.popcenter.org/library/crimeprevention/volume_10/08-Ditto_Short.pdf.
Accessed October 8, 2013.
Elliot, D. S. (1995). Lies, Damn Lies and Arrest Statistics. The Sutherland Award
Presentation. The American Society of Criminology Meetings. Boston, MA. Available
at: http://www.colorado.edu/cspv/publications/papers/CSPV-015.pdf. Accessed
October 8, 2013.
Ferenbok, J. and Clement, A. (2012). Hidden Changes: from CCTV to “Smart” video
surveillance. In: A. Doyle, R. Lippert and D. Lyon, (Eds). Eyes Everywhere: The Global
Growth of Camera Surveillance. Routledge.
Gill, M.; Spriggs, A.; Little, R.; and Collins, K. (2006). What do murderers think about
the effectiveness of CCTV? Journal of Security Education, 2, (1), 11 – 17.
Gill, M. and Spriggs, A. (2005), Assessing the impact of CCTV, Home
Office
Research
Study,
no.
292,
London:
Home
Office.
Grandmaison, Rachel, and Pierre Tremblay. (1997). Évaluation des Effets de la Télé—
Surveillance sur la Criminalité Commise dans 13 Stations du Métro de Montréal.
Criminologie 30:93-110.
Griffiths, M. (2003). Town centre CCTV: An examination of crime reduction in
Gillingham, Kent. Reading, UK: University of Reading. Available at:
http://www.popcenter.org/Responses/video_surveillance/PDFs/Griffith_nd.pdf.
Accessed October 8, 2013.
Groombridge, N. (2008), Stars of CCTV? How the Home Office wasted
millions – a radical ‘Treasury/Audit Commission’ view, Surveillance and
Society, 5(1), 73-80.
Haering, N., Venetianer, P., and Lipton, A. (2008) The evolution of video surveillance:
an overview. Machine Vision and Applications 19, 279–290.
Heilmann, E. (2003). La vidéosurveillance, une réponse à la criminalité?.
Criminologie,
vol.
36,
n.
1,
89-102.
Available
at
http://www.erudit.org/revue/crimino/2003/v36/n1/006554ar.html.
Accessed
October 8, 2013.
Huff, D. (1954) How to Lie with Statistics. New York: Norton.
Introna, L. D. and Wood, D. (2004). Picturing Algorithmic Surveillance: The Politics of
Facial Recognition Systems. Surveillance and Society. CCTV Special (eds. Norris,
McCahill and Wood) 2(2/3): 177-198. Available at: http://www.surveillance-andsociety.org/cctv.htm. Accessed October 8, 2013.
32
Introna, L. D. and Nissenbaum, H. (2009). Facial Recognition Technology: A Survey of
Policy and Implementation Issues. July, 22. Center for Catastrophe Preparedness and
Response,
New
York
University.
Available
at
http://www.nyu.edu/ccpr/pubs/Niss_04.08.09.pdf. Accessed October 8, 2013.
IRISS (2012). Surveillance, Fighting Crime and Violence. Report of the project
Increasing Resilience in Surveillance Societies. Available at: http://irissproject.eu/wpcontent/uploads/2012/02/IRISS_D1_MASTER_DOCUMENT_17Dec20121.pdf.
Accessed October 8, 2013.
Justice Analytical Services (2009), The Effectiveness of Public Space
CCTV: A review of recent published evidence regarding the impact of
CCTV on crime, Police and Community Safety Directorate, Edinburgh: The
Scottish
Government.
Available
at:
http://www.scotland.gov.uk/Resource/Doc/294462/0090979.pdf. Accessed October
8, 2013.
Maguire, M. (2002). Criminal Statistics: The ‘Data Explosion’ and its Implications in
M. Maguire, R. Morgan and R. Reiner (eds.). The Oxford Handbook of Criminology,
Third Edition. Oxford: Oxford University Press.
Mayhew, P., Clarke, R. V. G., Burrows, J. N., Hough, J.M. and Winchester, S. W. C.
(1979) Crime in public view. Home Office Research study 49. London: HMSO
Massumi, B. (2007). Potential Politics and the Primacy of Preemption. Theory and
Event
10:2.
Available
at:
http://muse.jhu.edu/login?auth=0andtype=summaryandurl=/journals/theory_and_e
vent/v010/10.2massumi.html. Accessed October 8, 2013.
Mayer-Schönberger, V. (2009). Delete: The Virtue of Forgetting in the Digital Age.
Princeton: Princeton University Press.
Melgaço, L. (2012). Estudantes sobre controle: a racionalização do espaço escolar
através do uso de câmeras de vigilância. O Social em Questão, year XIV, n. 27, 193212.
Available
at:
http://osocialemquestao.ser.pucrio.br/media/OSocial27_Se%C3%A7%C3%A3o_Livre_Melga%C3%A7o1.pdf. Accessed
October 8, 2013.
Monmonier, M. (1996) How to lie With Maps. 2nd ed. Chicago and London: University
of Chicago Press.
Morin, E. (2005). Introduction à la pensée complexe. Lonrai-França: Éditions du Seuil.
Norris, C. and Armstrong, G. (1999). The maximum surveillance society: The rise of
CCTV. Oxford, UK: Berg.
Owen, K. et al. (2006), A short evaluation of the (economic) benefits
of the Milton Keynes CCTV system in managing police resources,
Leicester: Perpetuity Research and Consultancy International (PRCI)
Ltd.
33
Reiner, R. (2002) Media made criminality: The representation of crime in the mass
media. In: Reiner, Robert, Maguire, Mike and Morgan, Rod, (eds.) The Oxford
Handbook of Criminology. Oxford University Press, Oxford, UK, 302-340.
Rios-Cabrera, R., Tuytelaars, T., and Van Gool, L. (2012). Efficient multi-camera
vehicle detection, tracking, and identification in a tunnel surveillance application.
Computer Vision and Image Understanding 116, 742–753.
Sivarajasingham, V.; Shephard, J.P.; and Matthews, K. (2003). Effect of urban closed
circuit television on assault injury and violence detection. Injury Prevention, 9, (4),
312 – 316.
Smith, G. (2004). Behind the Screens: Examining Constructions of Deviance and
Informal Practices among CCTV Control Room Operators in the UK, Surveillance and
Society,
2(2/3):
376-395.
Available
at:
http://www.surveillance-andsociety.org/articles2(2)/screens.pdf. Accessed October 8, 2013.
Stirling, A. (2010). Keep it complex. Nature, 468, 1029–1031. 23 December.
Waples, S., Gill, M. and Fisher, P. (2009), Does CCTV displace crime?,
Criminology and Criminal Justice, 9(2): 207-224.
Webb, B. and Laycock, G. (1992). Reducing crime on the London Underground: An
evaluation of three pilot projects. Crime Prevention Unit Series paper no. 30.
London: Home Office. Available at: http://www.popcenter.org/library/scp/pdf/189Webb_and_Laycock.pdf. Accessed October 8, 2013.
Welsh, B. C. and Farrington, D. P. (2009). Public Area CCTV and Crime Prevention: An
Updated Systematic Review and Meta‐Analysis. Justice Quarterly 26(4), 716-745.
Welsh, B. C. and Farrington, D. P. (2007). Closed-circuit television surveillance and
crime prevention: A systematic review. Stockholm: National Council for Crime
Prevention.
Welsh, B. C. and Farrington, D. P. (2002). Effects of Closed-Circuit Television on
Crime. The ANNALS of the American Academy of Political and Social Science 2003
587: 110-135.
i
"Alors que la pensée simplifiante désintègre la complexité du réel, la pensée complexe intègre le plus
possible les modes simplifiants de penser, mais refuse les conséquences mutilantes, réductrices,
unidimensionnalisantes et finalement aveuglantes d’une simplification qui se prend pour le reflet de
ce qu’il y a de réel dans la réalité." (Morin, 2005, p. 11).
ii
For an analysis of costs and benefits of other surveillance technologies see IRISS (2012).
iii
For the distinction between data, information, knowledge and wisdom, see Ackoff (1989).
34